def __init__(self, config=None, env=None, logger_creator=None):
"""Initialize an RLLib agent.
Args:
config (dict): Algorithm-specific configuration data.
env (str): Name of the environment to use. Note that this can also
be specified as the `env` key in config.
logger_creator (func): Function that creates a ray.tune.Logger
object. If unspecified, a default logger is created.
"""
config = config or {}
# Agents allow env ids to be passed directly to the constructor.
self._env_id = env or config.get("env")
Trainable.__init__(self, config, logger_creator)
def __init__(self, config=None, env=None, logger_creator=None):
"""Initialize an RLLib trainer.
Args:
config (dict): Algorithm-specific configuration data.
env (str): Name of the environment to use. Note that this can also
be specified as the `env` key in config.
logger_creator (func): Function that creates a ray.tune.Logger
object. If unspecified, a default logger is created.
"""
config = config or {}
if tf and config.get("eager"):
tf.enable_eager_execution()
logger.info("Executing eagerly, with eager_tracing={}".format(
"True" if config.get("eager_tracing") else "False"))
if tf and not tf.executing_eagerly():
logger.info("Tip: set 'eager': true or the --eager flag to enable "
"TensorFlow eager execution")
# Vars to synchronize to workers on each train call
self.global_vars = {"timestep": 0}
# Trainers allow env ids to be passed directly to the constructor.
self._env_id = self._register_if_needed(env or config.get("env"))
# Create a default logger creator if no logger_creator is specified
if logger_creator is None:
timestr = datetime.today().strftime("%Y-%m-%d_%H-%M-%S")
logdir_prefix = "{}_{}_{}".format(self._name, self._env_id,
timestr)
def default_logger_creator(config):
"""Creates a Unified logger with a default logdir prefix
containing the agent name and the env id
"""
if not os.path.exists(DEFAULT_RESULTS_DIR):
os.makedirs(DEFAULT_RESULTS_DIR)
logdir = tempfile.mkdtemp(
prefix=logdir_prefix, dir=DEFAULT_RESULTS_DIR)
return UnifiedLogger(config, logdir, loggers=None)
logger_creator = default_logger_creator
Trainable.__init__(self, config, logger_creator)
def __init__(
self, config={}, env=None, registry=get_registry(),
logger_creator=None):
"""Initialize an RLLib agent.
Args:
config (dict): Algorithm-specific configuration data.
env (str): Name of the environment to use. Note that this can also
be specified as the `env` key in config.
registry (obj): Object registry for user-defined envs, models, etc.
If unspecified, the default registry will be used.
logger_creator (func): Function that creates a ray.tune.Logger
object. If unspecified, a default logger is created.
"""
# Agents allow env ids to be passed directly to the constructor.
self._env_id = env or config.get("env")
Trainable.__init__(self, config, registry, logger_creator)
def __init__(
self, config=None, env=None, registry=None,
logger_creator=None):
"""Initialize an RLLib agent.
Args:
config (dict): Algorithm-specific configuration data.
env (str): Name of the environment to use. Note that this can also
be specified as the `env` key in config.
registry (obj): Object registry for user-defined envs, models, etc.
If unspecified, the default registry will be used.
logger_creator (func): Function that creates a ray.tune.Logger
object. If unspecified, a default logger is created.
"""
config = config or {}
# Agents allow env ids to be passed directly to the constructor.
self._env_id = env or config.get("env")
Trainable.__init__(self, config, registry, logger_creator)
def __init__(self, config=None, env=None, logger_creator=None):
"""Initialize an RLLib agent.
Args:
config (dict): Algorithm-specific configuration data.
env (str): Name of the environment to use. Note that this can also
be specified as the `env` key in config.
logger_creator (func): Function that creates a ray.tune.Logger
object. If unspecified, a default logger is created.
"""
config = config or {}
Agent._validate_config(config)
# Vars to synchronize to evaluators on each train call
self.global_vars = {"timestep": 0}
# Agents allow env ids to be passed directly to the constructor.
self._env_id = env or config.get("env")
# Create a default logger creator if no logger_creator is specified
if logger_creator is None:
timestr = datetime.today().strftime("%Y-%m-%d_%H-%M-%S")
logdir_prefix = "{}_{}_{}".format(self._agent_name, self._env_id,
timestr)
def default_logger_creator(config):
"""Creates a Unified logger with a default logdir prefix
containing the agent name and the env id
"""
if not os.path.exists(DEFAULT_RESULTS_DIR):
os.makedirs(DEFAULT_RESULTS_DIR)
logdir = tempfile.mkdtemp(prefix=logdir_prefix,
dir=DEFAULT_RESULTS_DIR)
return UnifiedLogger(config, logdir, None)
logger_creator = default_logger_creator
Trainable.__init__(self, config, logger_creator)
def __init__(self, config=None, env=None, logger_creator=None):
"""Initialize an RLLib agent.
Args:
config (dict): Algorithm-specific configuration data.
env (str): Name of the environment to use. Note that this can also
be specified as the `env` key in config.
logger_creator (func): Function that creates a ray.tune.Logger
object. If unspecified, a default logger is created.
"""
config = config or {}
# Vars to synchronize to evaluators on each train call
self.global_vars = {"timestep": 0}
# Agents allow env ids to be passed directly to the constructor.
self._env_id = self._register_if_needed(env or config.get("env"))
# Create a default logger creator if no logger_creator is specified
if logger_creator is None:
timestr = datetime.today().strftime("%Y-%m-%d_%H-%M-%S")
logdir_prefix = "{}_{}_{}".format(self._agent_name, self._env_id,
timestr)
def default_logger_creator(config):
"""Creates a Unified logger with a default logdir prefix
containing the agent name and the env id
"""
if not os.path.exists(DEFAULT_RESULTS_DIR):
os.makedirs(DEFAULT_RESULTS_DIR)
logdir = tempfile.mkdtemp(
prefix=logdir_prefix, dir=DEFAULT_RESULTS_DIR)
return UnifiedLogger(config, logdir, None)
logger_creator = default_logger_creator
Trainable.__init__(self, config, logger_creator)